This chapter presents a discussion of the approaches toward learning protected biometric templates using neural networks. Learned biometric template protection (BTP) methods are different to the handcrafted techniques from Chaps. 4 – 6 of this book, in that the handcrafted algorithms are explicitly defined by humans, while the learned approaches rely on neural networks to learn how to generate a protected template from its unprotected version. Learned BTP methods have been studied mainly for face template protection, so this will be the primary context in which these methods will be investigated in this chapter; however, the presented techniques are by no means exclusive to the face modality. The chapter consists of three main sections, which build upon each other to track the development of Learned BTP research. The first section introduces the earliest approaches, which train a neural network to learn a mapping from the biometric template to a predefined (random) code. The second section explores the move toward a more flexible approach, where neural networks are trained to learn their own representation of a protected template instead of learning a mapping to a predefined code. The third section discusses the most recent Learned BTP approaches, which incorporate additional randomness into the learning process to allow for the renewal of the learned protected templates in the event of compromise.